So, I’m in school for data science. I see AI & ML applications for film scanning. So my school project is a film scanner. The twist is that I’m going back to school 20+ years after going to college, and having worked on everything from the first “touch/no touch” screens, to analog retro tech. I also was a “maker” teacher and co-ordinator, so yeah, I have a laser, 3d printer, machine shop, and an electronics lab that’d be able to handle new tech.
I’ve already gone through the gauntlet of professional and casual users research as per my requirements.
Cost must be kept as low as possible, but I can use any reasonable trick I want to do so. I.e. if you already have a video editing rig, you therefore have a high end gpu.
Speed is not a factor, as long as it’s faster than the animal named competition. And of course I must use data analytics, AI, ML, etc.
A good proportion of “main” problems with the kinograph are already solved on paper, and in some cases in practice with my prototype.
But, here’s the other twist! If I do a theory only project, I get massively less points. So I’m highly motivated to do a “real” build of many units.
Now I’ve already found the hard limits & want to work through them.
I’ve gone through the research up to usb4 & “other” interfaces. I skip everything else as it blows the budget beyond comprehension.
I now understand why there’s a jump from $200, to $2000, to $20k+. Although I understand that middle category may no longer exist.
Tech specs for those that are curious, 12MP sensor with synth HDR(4096x3000ish). Stepper drives run off smooth drive CNC controllers. Arduino main PC/Mac/Linux interface to straight hardware. Camera is usb3-4, or other.(yes it’s flexible). Possible PC less operation with a RaspPi 4 type unit, but I’m considering better options. For magnetic audio, it’ll have to be a second pass with a crystal lock analog-ish system(8mm/super8 audio is magnetic). Lens is the cheapest garbage, but it’s better than high end lenses due to basic math errors.
Current price point $1000.
Can you elaborate? The lens is a rather critical component for a good film scanner…
What I did was collect lenses of every kind I could find, video, 35mm, projector, c-mount, microscope, document scanner, and rare lenses. Made the simplest rig with ground glass and just observed and measured all the details of each method of imaging. Then who built the lens & why.
It boiled down to not “good” glass, or even cost. Better working lenses were often remanufactered from another format. So better functioning C-mount lenses were 16mm ones originally.
Another observation was the closer to 1:1 subject to sensor ratio, the less critical things get. Aha! Lightbulb moment. This is why an “ok” to “almost bad” lens used with a macro extension tube can be a “good” to “excellent” lens.
The last piece of the puzzle was what tricks do higher end document scanners and their optics use. Pretty simple lenses in most cases, definitely not german made. But they all had baffles on or near the sensor. Ah, they’re using a larger image circle than the sensor and cutting off any out of focus or distorted region.
So I tried an on camera experiment. C mount type macro lens for a 1/3" ccd vs “5MP super zoom” originally for a 1/2" sensor on a spacer. The “right lens” had distortion and light fall off in 4 corners, and slight chromatic issues as well. Exactly the problem others describe. Although the focus distance was great. The “wrong” lens had almost none of the problems. With a bit of zooming and adjusting spacing nearly 1:1 equavalent ratio can be achieved. Short of a telecentric, it’s usable without calibration or digital image correction.
Please understand I’m doing 8mm sized film and 1/3"(ish) sensors. But the things I’ve learned are applicable to larger film and sensors.
Interesting im embarking on building my 8mmStd/Super scanner fortunately not require sound.
I’d be interested to hear more about your lens selection as im working on that now and im looking at purchasing one of these cameras www.arducam.com then their Pivariety series either 16M or 21M Global shutter version.
They are full color and output in RAW format and reasonably priced $100usd for the 21M version. 16M uses the Sony IMX298 sensor and the 21M uses the Sony IMX230 sensor.
Both hook straight into the RPi and they have software drivers already sorted.
Thoughts???
Sounds like a great collab. Go ahead and find me on facebook for instant messenger. Yea the arducams have a lot going for them. Especially in the realm of straight to PI functionality.
However, those referred to units are both -->modules<-- effectively a finished lens camera solution. I have a few similar units to see how to undo the integrated lens.
If the sensor data talks about autofocus, then the lens is unfortunately integrated. I’d absolutely love a sensor with only the liquid focuser, and no lens. But being an ex Sony tech, I’m 99% certain that isn’t easily possible.
Not all is lost however, it’s possible to use human eye type optics in front of such a sensor. I.e. a long focus microscope can be used with a phone quite well.
This link is a great resource for lens information. Many are using the Schneider Componon-S 2.8 and the specs are probably the best in that range. I had and use the Nikkor EL 50mm 2.8N -new version, good resolution for 8 and Super 8. Presently using a DSLR D3200 and the lens keeps up.